Top Banner
The Ensemble-Building Challenge for Fashion Recommendation: Investigation of In-home Practices and Assessment of Garment Combinations Jingwen Zhang, Loren Terveen, and Lucy E. Dunne University of Minnesota
20

The Ensemble-Building Challenge for Fashion Recommendation

Feb 01, 2022

Download

Documents

dariahiddleston
Welcome message from author
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Page 1: The Ensemble-Building Challenge for Fashion Recommendation

The Ensemble-Building Challenge for Fashion Recommendation:

Investigation of In-home Practices and Assessment of Garment Combinations

Jingwen Zhang, Loren Terveen, and Lucy E. DunneUniversity of Minnesota

Page 2: The Ensemble-Building Challenge for Fashion Recommendation

Hypothetical woman’s closet of 120 items:• 15 dresses• 25 bottoms (pants/skirts)• 30 tops• 30 sweaters• 20 jackets

Total outfits?

25*30*30*20+15 = 675,015 … nothing?

I have nothing to wear!

Page 3: The Ensemble-Building Challenge for Fashion Recommendation

Understanding Outfits

Self-reported objectives and

challenges

Self-reported strategies

Assessment of auto-generated

outfits

Page 4: The Ensemble-Building Challenge for Fashion Recommendation

Self-report: User survey

194 MTurk workers, 18-63 yrs old, 128 female/66 male

“Consumer spectrum score” calculated from a 9-question battery

Survey topics:(1) Perceived wardrobe size and use(2) Values and objectives in dressing(3) Constraints of the wardrobe moment(4) Variables at play in the dressing decision(5) Outfit-building strategies

Page 5: The Ensemble-Building Challenge for Fashion Recommendation

Results: Wardrobe Size

Larger wardrobes for female vs. male participants

Slightly larger wardrobes when estimated by sub-category

No strong relationships between wardrobe size and consumer spectrum score

-10

0

10

20

30

40

50

60

70

80

90

Male Female

Self-reported Working Wardrobe Size

Estimated overall Estimated by category

Page 6: The Ensemble-Building Challenge for Fashion Recommendation

Results: Wardrobe Use

Percent of working wardrobe in “regular use”

Female Male

In use (63%)

In use (58%)

Page 7: The Ensemble-Building Challenge for Fashion Recommendation

Objectives in Dressing

1. I want to look my best or improve the way I look2. I want to make use of everything I own and waste as little as possible3. I want to look trendy and keep up with current fashion4. I want to have fun putting together outfits, be creative, and express myself5. I want to fit in and look appropriate every day6. I want to dress to flatter my body7. I want to reduce my consumption of clothing8. I want to be comfortable (physically) in my clothes every day9. I want to look unique and different in the way I dress10. I want to get dressed as quickly as possible11. I want to spend less money on clothing12. I want to do laundry less often

Page 8: The Ensemble-Building Challenge for Fashion Recommendation

Results: Objectives in Dressing

1. Look my best 2. Make use of everything3. Look trendy 4. Have fun 5. Fit in and look appropriate6. Flatter my body7. Reduce consumption 8. Be comfortable9. Look unique 10. Dress quickly11. Spend less money 12. Do laundry less often

Page 9: The Ensemble-Building Challenge for Fashion Recommendation

Results: Objectives in Dressing

Page 10: The Ensemble-Building Challenge for Fashion Recommendation

The Wardrobe Moment

Male participants: 61.54% spend less than 5 min dressing, 95.39% less than 10 min

Female participants: 42.97% spend less than 5 min, 80.47% less than 10 min

For a “special day”, 38.46% male/60.94% female spend more than 10 min

Page 11: The Ensemble-Building Challenge for Fashion Recommendation

Sources of Difficulty in Dressing

Page 12: The Ensemble-Building Challenge for Fashion Recommendation

Strategies in Dressing

Page 13: The Ensemble-Building Challenge for Fashion Recommendation

Strategies in Dressing

Page 14: The Ensemble-Building Challenge for Fashion Recommendation

Strategies in Dressing: Outfit-Building

Starting point for outfit-building: pick a garment

Use a constraint: weather, activity, what’s clean

Make an emotional choice: what color do you feel like today?

Page 15: The Ensemble-Building Challenge for Fashion Recommendation

Evaluating Outfits

Page 16: The Ensemble-Building Challenge for Fashion Recommendation

Test Wardrobe and Outfits

137-item wardrobe: 77 tops, 10 bottoms, 10 dresses, and 5 jackets

Tops categorized in 3 layers (inner, middle, outer)

491,185 one- to three-layer outfits generated

Page 17: The Ensemble-Building Challenge for Fashion Recommendation

Rating Outfits

Random sample of 1000 outfits extracted, rated on a 5-point scale:

5: This is a great outfit; I can imagine someone looking good wearing exactly this.4: This is an ok outfit. It might have some style problems, or it might be a little bland, but it's wearable.3. This is a wearable outfit, but it has some problems. These garments could technically be worn together, but the outfit doesn't work very well. 2: This outfit has serious problems, it would be hard to imagine someone wearing it, but a few people might.1: This outfit is not wearable; I can't imagine anyone wearing it in public.

Each outfit rated by 3 “expert” members of the research team, plus 2-4 crowdsourced raters

Page 18: The Ensemble-Building Challenge for Fashion Recommendation

Outfit Rating All Raters

ExpertRaters

Lay Raters

>=3 (Could technically be worn together)

90.1% 91.6% 86.0%

>=4 (An “ok” outfit) 61.0% 72.5% 48.6%

=5 (A “great” outfit) 1.2% 34.0% 2.2%

Results: Rating Outfits

Page 19: The Ensemble-Building Challenge for Fashion Recommendation

Conclusions

Individuals self-report smaller wardrobes that are in more regular use than other evidence indicates

The “wardrobe moment” is a short-term (~5 min) interaction, decisions are constrained by dressing objectives as well as practical considerations

Objectives and strategies in dressing vary across individuals, some variables are linked to position on the consumer spectrum. There is more consistency in goals like wanting to be comfortable, look better, and use the wardrobe efficiently.

Recommending an outfit should take into account practical considerations like weather and activity as well as subjective factors like comfort and social appropriateness. Trendiness was not prioritized.

Individuals are unlikely to perceive having “too many options” as a source of difficulty in dressing, yet they have trouble finding “good” options and the vast majority of garment combinations make wearable outfits. Further research is needed to measure and predict outfit quality for a specific user.

Page 20: The Ensemble-Building Challenge for Fashion Recommendation

Thanks! Questions?

[email protected]@[email protected]